📑 Feature Request: Pre-commit hooks for sanity checking

[DhanshreeA]

Is your feature request related to a problem? Please describe.

For a successful contribution in the Ersilia model hub, there are a couple of routine tasks that the contributor has to perform. For example,

• When adding model checkpoints, they have to be tracked using Git LFS instead of files simply tracked by Git. In this case, contributors are likely to forget tracking these large files with Git LFS.
• Updating metadata.json such that the README gets generated automatically. Right now, there is a GitHub Actions workflow that parses the metadata.json to check for inconsistencies before a contributor can successfully create a PR. However, this is increasing time to delivery because contributors push their changes, create a PR and move away to a different task, sometimes missing whether all workflows on PR creation ran successfully or not.

Describe the solution you'd like.

We can use pre-commit hooks in model repositories such that automated checks are made even before a contributor is able to commit code on their local fork. These git hooks can help us with a lot of things, but primarily we can prevent

• Tracking of large files within Git instead of Git LFS using check-added-large-files
• Enforce valid JSONs before code commits using check-json

We can potentially even prettify and enforce code quality if needed, however for now I think this could help us resolve problems we encounter during PRs even before they are made.

Describe alternatives you've considered

No response

Additional context.

No response

[GemmaTuron]

Hi @DhanshreeA !

I've never done pre-commit hooks, do you know how to? The main issue I foresee is that we want to test that the metadata is fine and that the model is working, and all this is impossible to do in the fork I'm afraid, it needs to be run from the Ersilia organisation repositories, hence at PR time.

[GemmaTuron]

Follow up on this:

• There is a json syntax check that will look for syntax errors in the metadata file first
• The README gets automatically updated once the PR is merged. Contributors MUST wait for the PR to pass all the checks before moving on
• The git-lfs issue: is it worth adding a check for that? it is complicated to work with Git-LFS so I am kind of hesitant to add another check for that, but happy to hear your suggestion @DhanshreeA

[DhanshreeA]

Hi @GemmaTuron first of all apologies for abandoning this issue. I'll create a quick POC to see if we can use this and share it here, otherwise we can let this go for now. I'll update here by the end of the week.